Real-time object detection system on PAC Duo platform

State-of-the-art object detection methods are usually developed on a PC-based platform in consideration of its implementation feasibility. However, various applications such as car license recognition, lane departure warning system or telecare system require the use of an embedded platform. In this paper, we proposed approaches to migrate object detection algorithm while realizing it on an embedded system. The accomplishment takes advantage of both software parallelism and hardware feature of PACDSP processor to fully utilize its computational resource. These ideas can also apply on other image processing applications with similar operations to largely improve system performance. Experimental result shows that PAC Duo platform can correctly detect up to 10 objects in real-time at a significantly lower DSP working frequency of 150 MHz. Furthermore, only 18 KB of program memory and 19.6 KB of internal data memory in the embedded system are used, which reduces the cost of the system comparing to other works.